[foreign-memaccess+abi] RFR: 8263018: Improve API for lifecycle of native resources [v10]

[Maurizio Cimadamore]

On Mon, 15 Mar 2021 17:18:37 GMT, Paul Sandoz <psandoz at openjdk.org> wrote:

> > A possible, improved naming for the two segment allocators:
> > SegmentAllocator.of(MemorySegment) -> SegmentAllocator.slicing(MemorySegment)
> > SegmentAllocator.of(ResourceScope) -> SegmentAllocator.scoped(ResourceScope)
> > What do you think?
> 
> Yes, that's better. I find it hard to come up with a better concise name for the former, allocation returns a sliced prefix of the given segment, thus allocations are aliased, which is fundamentally different to all the other cases.

One option which I still find kind of appealing, is to add _instance_ methods to connect the various APIs. E.g. instead:

 SegmentAllocator.of(scope) -> scope.toAllocator();
 SegmentAllocator.of(segment) -> segment.toSegment();
 
 I think these conversions are great in certain contexts: think of a client slicing a segment and wanting to put the result of a native call in a certain place:

MemorySegment segment = ...
MemorySegment out = segment.asSlice(...);
linkerHandle.invokeExact(out.toAllocator(), ...);

Similarly, if you have an resource scope and simply want an allocator that returns segments attached to that scope:

try (ResourceScope scope = ...) {
   linkerHandle.invokeExact(scope.toAllocator(), ...);
}

I find both more concise that their static variants, but also (crucially) easier to explain. What we lose if we go down that path is "independence" between the various abstractions - in the sense that a ResourceScope is no longer Panama-neutral.

-------------

PR: https://git.openjdk.java.net/panama-foreign/pull/466